An opinionated starter project for Model Context Protocol (MCP) servers
This is a starter template with a single example tool that demonstrates basic structure but lacks production-grade quality across multiple dimensions. The tool 'some_function' has a vague description, minimally compliant schema, and no error handling guidance. The template itself is well-scaffolded (good build setup, linting, test structure) but the actual tool definition falls short of baseline standards. The description 'An example tool' is perfunctory and provides no context for when or why an LLM should invoke this tool. The parameter descriptions are present but superficial ('Name of the person', 'Address object'). There is no documentation of output structure, no error recovery hints, and no guidance on expected behavior.
An example tool
Tool name 'some_function' is generic and does not start with an action verb. It does not communicate what the tool does, is it a lookup, a mutation, a computation?
Tool description 'An example tool' is 16 characters, below the 10 - 1024 character baseline and far too vague. LLMs cannot determine when to select this tool vs alternatives. Should explain: what does it do, when to call it, what it returns.
No output schema is documented. The tool description and handler do not specify what the response structure is (fields, types, cardinality). Agents have no way to plan downstream tool calls.
Address parameter uses a nested object with required fields ['street', 'zip'] but the description 'Address object' does not explain format expectations, validation, or what happens if fields are missing.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 40 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 22 | - | v1 |
No error handling guidance. The tool declares Risk: READ_ONLY, but there is no documentation of failure modes, invalid inputs, or recovery steps. An LLM encountering a failure has no guidance on what to do next.
Parameter 'name' is described as 'Name of the person' but does not specify format, length, character restrictions, or whether it accepts email vs full name vs username. Too vague for LLM to construct valid inputs reliably.